Platform
ScaiWave ScaiGrid ScaiCore ScaiBot ScaiDrive ScaiKey Models Tools & Services
Solutions
Organisations Developers Internet Service Providers Managed Service Providers AI-in-a-Box
Resources
Support Documentation Blog Downloads
Company
About Research Careers Investment Opportunities Contact
Log in

Changelog

0.1.0 — first public release (2026-05-15)#

The initial public ship. ScaiFlux launches as a self-contained Linux binary (no Python, no pip, no virtualenv on the target machine) plus a standard wheel for pip / pipx users.

What you get#

  • Interactive REPL (scaiflux repl or just scaiflux) — tab completion for slash commands, live model output streaming, session persistence, ESC to interrupt a turn, on-screen context-window usage indicator.
  • One-shot mode (scaiflux prompt "...") — pipe-friendly, emits text or JSON.
  • Serve mode (scaiflux serve) — NDJSON-over-stdio protocol so external orchestrators (ScaiForge etc.) can drive a sandboxed ScaiFlux process with full interactivity and proxied permissions. See serve protocol reference.
  • First-run wizard (scaiflux setup) — three-step interactive config: ScaiGrid endpoint, API key, default model. The REPL detects first-run and offers to dispatch the wizard automatically.
  • Built-in tools — Bash, ReadFile, WriteFile, EditFile, GlobSearch, GrepSearch (ripgrep-backed when available), WebFetch, WebSearch, TodoWrite, NotebookEdit, PdfExtract, Skill, Agent (sub-agents), TaskRun, the task/team/cron registries, ToolSearch.
  • Per-workspace config via .scaiflux.json (JSON-schema validated) and project memory via SCAIFLUX.md.
  • MCP client support (install scaiflux[mcp] for the wheel; the standalone binary doesn't include it).
  • Permission modes: read-only, workspace-write, danger-full-access, prompt, plan.
  • Auto-retry for transient ScaiGrid errors (3 attempts, exponential backoff, only on remote-classified failures, never after a partial stream).

Get it#

Known limits#

  • The standalone binary is Linux x86_64 only this release. macOS, Windows, and ARM64 builds are on the roadmap; in the meantime pip install works on every platform Python supports.
  • Entry-point plugins (third-party tools registered via [project.entry-points."scaiflux.tools"]) aren't picked up by the standalone binary — there's no pip install path inside a frozen executable. Plugin authors should ship the wheel.
  • MCP client support is wheel-only for the same reason.
Updated 2026-05-18 14:34:14 View source (.md) rev 2